AWS Radar Architecture
A showcase of AWS serverless services working together to build an intelligent, scalable, cost-effective application.
Architecture Overview
AWS Services
Compute
AWS Lambda — Runs all application logic: fetching RSS, AI processing, API handlers. Python 3.13 on Graviton2 (arm64) for 20% cost savings. Scales to zero when idle.
AI/ML
Amazon Bedrock Nova 2 Lite — The brain of AWS Radar. Generates summaries, suggests topic tags, classifies AWS services. Fast, cheap, fully managed—no GPUs to provision.
Amazon Bedrock Titan Embeddings — Converts text to vectors for semantic matching. Enables fuzzy tag matching ("serverless" finds "lambda functions"). Stored directly in DynamoDB, no vector database needed.
Data
Amazon DynamoDB — Stores announcements, tags, services, and user preferences across 5 tables. On-demand billing means zero cost at zero traffic. Single-digit millisecond latency.
Messaging
Amazon SQS — Decouples RSS fetching from AI processing. Batch consumption (10 messages/invocation) reduces Lambda costs. Dead Letter Queue captures failures for debugging.
API & Frontend
Amazon API Gateway — REST API with Cognito authorization and response caching. Built-in throttling protects backend from traffic spikes.
Amazon CloudFront — Global CDN serving static frontend from edge locations. Single domain for frontend + API eliminates CORS complexity. Free HTTPS via ACM.
Amazon S3 — Hosts the Bootstrap frontend. No compute cost, just storage.
Authentication
Amazon Cognito — Managed auth with email/password and Google OAuth. Native API Gateway integration. First 50k monthly users free.
DNS & Certificates
Amazon Route 53 — DNS hosting with automatic CloudFront alias records.
AWS Certificate Manager — Free SSL/TLS certificates with automatic renewal.
Monitoring
Amazon CloudWatch — Metrics, logs, alarms, and dashboards. Custom metrics track announcements fetched/processed/failed. Alarms notify on DLQ messages or missing fetches.
AWS X-Ray — Distributed tracing from API Gateway through Lambda to DynamoDB and Bedrock. Finds latency bottlenecks.
CI/CD
AWS CodePipeline + CodeBuild — Automated GitHub-to-production deployment. Runs tests, builds with SAM, deploys via CloudFormation. Separate pipelines for backend and frontend.
Design Principles
1. Scale to Zero
Every service chosen supports zero-cost idle state:
- Lambda: Pay per invocation
- DynamoDB: Pay per request
- API Gateway: Pay per request
- SQS: Pay per message
- CloudFront: Pay per request
Result: Near-zero cost during low traffic.
2. Event-Driven Architecture
Clear separation of concerns:
Schedule → Fetch → Queue → Process → Store → Serve
Each stage independent, retryable, observable.
3. Managed AI
No ML infrastructure: No model hosting, no GPU provisioning, no training pipelines. Just API calls to Bedrock.
4. Security by Default
- All traffic HTTPS
- Cognito authentication
- IAM least privilege
- Secrets in Secrets Manager
- Encryption at rest (DynamoDB, S3, SQS)
5. Infrastructure as Code
Everything defined in YAML:
template.yaml— SAM (Lambda, API Gateway, CloudFront)infrastructure.yaml— DynamoDB, SQS, S3, Cognitopipeline.yaml— CI/CD
Reproducible, version-controlled, auditable.
Cost Profile
| Service | Cost Model | Typical Monthly |
|---|---|---|
| Lambda | Per invocation | ~$1 |
| DynamoDB | Per request | ~$1 |
| Bedrock | Per token | ~$5 |
| API Gateway | Per request | ~$1 |
| CloudFront | Per request | ~$1 |
| S3 | Storage | ~$0.10 |
| SQS | Per message | ~$0.10 |
| Cognito | Per MAU (free tier) | $0 |
| CloudWatch | Metrics + logs | ~$1 |
Total: ~$10/month for typical usage (scales with traffic).
Why This Architecture
AWS Radar demonstrates that you can build a production-grade, AI-powered application with:
- Zero servers to manage
- Zero capacity planning
- Near-zero idle cost
- Automatic scaling
- Built-in security
- Full observability
The serverless model lets you focus on features, not infrastructure.